In his lecture, he points out that economists seem to have come to believe that only that what can be measured belongs in models. As a result, the often immeasurable complexities that exist in the economies of countries are largely ignored. Ironically, this is done to be scientific in the application of thinking to the domain of economics. He refers to the concept of phenomena of organized complexity social sciences have to deal with to make this point, arguing it is rather unscientific to believe such complexity can be described through models capturing only those factors we can observe and measure.

Very explicitly he seems to warn against the idea we could manage inflation (a numerical constant) to be withing a target band (more numerical constants) by setting just 1 numerical constant every few months, as is common practice these days:

It has led to the illusion, however, that we can use this technique
for the determination and prediction of the numerical values of those
magnitudes; and this has led to a vain search for quantitative or
numerical constants. This happened in spite of the fact that the
modern founders of mathematical economics had no such illusions.

Remarkably, he even touches on the possibility of using computers to derive predictions from big multi-factor models, and warns these still won't provide numerical accuracy where the real difficulty is ascertaining the facts. (The example he provides might sound a bit dated to modern-day sports data scientists.)

Funny as that may be, the entire text can be read as a marvellous admonition of the idea for of allowing central banks to determine what effectively should be the interest rate citizens of a country get to pay on their mortgage based on the theory that inflation should be between 2 and 3% and people's mortgages are a key avenue to achieve that goal.